DYNAMIC PASSWORD VOICE BASED IDENTITY AUTHENTICATION SYSTEM AND METHOD HAVING SELF-LEARNING FUNCTION

• US 20170365259A1
• Filed: 07/08/2015
• Published: 12/21/2017
• Est. Priority Date: 02/05/2015
• Status: Active Grant

• Current Assignee: Beijing d-Ear Technologies Co., Ltd.
• Original Assignee: Beijing d-Ear Technologies Co., Ltd.
• Inventors: ZHENG, Fang, WU, Xiaojun, LI, Lantian, WANG, Gang, WANG, Xiaogang, LIU, Le
• Granted Patent: US 10,540,980 B2
• Time in Patent Office: Days
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 17/02 Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
• CPC Class Codes: G10L 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 17/24 the user being prompted to utter a password or a predefined phrase
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum
• CPC Class Codes: H04L 63/083 using passwords cryptographic mechanisms or cryptographic arrangements for entity authentication using a predetermined code H04L9/3226
• CPC Class Codes: H04L 63/0861 using biometrical features, e.g. fingerprint, retina-scan cryptographic mechanisms or cryptographic arrangements for entity authentication using biological data H04L9/3231
• CPC Class Codes: H04W 12/06 Authentication

Abstract

Dynamic security code speech-based identity authentication system and method having self-learning function, being equipped with: a time-varying data storage unit for storing speech data of each user with time labels; a time-varying data update module (23) for storing the latest speech data into the time-varying data storage unit; a time window channel construction module (24) for extracting speech data from the time-varying data storage unit in the order of the time labels, constructing and updating the time window channel comprising multiple sets of speech data; a voiceprint model reconstruction module (25) reconstructing the user voiceprint model by using the multiple sets of speech data comprised in the updated time window channel.

Claims

1. A dynamic security code speech-based identity authentication system having self-learning function, comprising:

a request receiving module for receiving an identity authentication request that a requester sends to a server through a client;

a dynamic security code generating module for generating a dynamic security code and sending the dynamic security code to the client; and

an identity authentication module for calculating a comprehensive confidence of an identity of the requester by using an acoustic model of global characters and a voiceprint model of a user based on a security code speech signal sent from the client, wherein the security code speech signal is generated when the requester reads out the dynamic security code;

judging the identity of the requester based on the calculated comprehensive confidence of the identity; and

feeding an identity authentication result back to the client, wherein

the dynamic security code speech-based identity authentication system is provided with an automatic reconstruction subsystem for the voiceprint model, and the voiceprint model of the user is reconstructed by the automatic reconstruction subsystem for the voiceprint model when the identity authentication result is that the requester is the user of the server, and

the automatic reconstruction subsystem for the voiceprint model comprises;

a time-varying data storage unit for storing speech data of each user with time labels;

a time-varying data updating module for storing the security code speech signal as a latest speech data into the time-varying data storage unit;

a time window channel construction module for extracting the speech data of the user from the time-varying data storage unit in an order of the time labels, constructing a time window channel including a plurality of sets of speech data, and updating the speech data included in the time window channel using the latest speech data; and

a voiceprint model reconstruction module for reconstructing the voiceprint model of the user for the user using the plurality of sets of speech data included in the updated time window channel.

2. The dynamic security code speech-based identity authentication system having self-learning function according to claim 1, wherein

the automatic reconstruction subsystem for the voiceprint model further comprises a parameterization module for speech data, and the parameterization module for speech data is used for parameterizing the security code speech signal, i.e., speech data, to obtain a latest parameterized speech data;

parameterized speech data of each user is stored with time labels in the time-varying data storage unit;

the latest parameterized speech data is stored in the time-varying data storage unit by the time-varying data updating module;

the time window channel construction module extracts parameterized speech data of the user from the time-varying data storage unit in the order of the time labels, constructs a time window channel including a plurality of sets of parameterized speech data, and updates the parameterized speech data included in the time window channel using the latest parameterized speech data; and

the voiceprint model reconstruction module reconstructs the voiceprint model of the user for the user using the plurality of sets of parameterized speech data included in the updated time window channel

3. The dynamic security code speech-based identity authentication system having self-learning function according to claim 2, wherein

the automatic reconstruction subsystem for the voiceprint model further comprises a speech recognition module for recognizing phonemes corresponding to respective frames in the speech data;

phonemes corresponding to the latest parameterized speech data and frame intervals corresponding to the phonemes are further stored in the time-varying data storage unit; and

the time window channel construction module updates the parameterized speech data included in the time window channel based on the phonemes corresponding to the latest parameterized speech data, so that phonemes corresponding to the plurality of sets of parameterized speech data included in the time window channel are evenly distributed.

4. The dynamic security code speech-based identity authentication system having self-learning function according to claim 3, wherein

the time window channel construction module compares the phonemes corresponding to the latest parameterized speech data with phonemes corresponding to an earliest parameterized speech data in the time window channel, and when a matching degree of the phonemes corresponding to the latest parameterized speech data and the phonemes corresponding to the earliest parameterized speech data is greater than or equal to a predetermined matching degree, the latest parameterized speech data is pushed into the time window channel, and the earliest parameterized speech data is removed from the time window channel.

5. The dynamic security code speech-based identity authentication system having self-learning function according to claim 3, wherein,

the time window channel construction module tentatively removes a set of parameterized speech data from the time window channel sequentially in the order of the time labels from old to new, and calculates an equilibrium degree of a character-based phoneme distribution based on all of parameterized speech data remained in the time window channel and the latest parameterized speech data, and if the equilibrium degree is greater than or equal to a predetermined threshold of the equilibrium degree, the latest parameterized speech data is pushed into the time window channel;

otherwise, the set of parameterized speech data tentatively removed is restored to the time window channel, and then a next set of parameterized speech data is tentatively removed from the time window channel, and once again, the time window channel construction module calculates the equilibrium degree based on all of parameterized speech data remained in the time window channel and the latest parameterized speech data till each set of parameterized speech data included in the time window channel has been tentatively removed or the latest parameterized speech data has been pushed into the time window channel.

7. The dynamic security code speech-based identity authentication system having self-learning function according to claim 1, wherein

the voiceprint model reconstruction module estimates an updating frequency for updating the speech data of the user stored in the time-varying data storage unit, and if the updating frequency is less than a predetermined frequency, the voiceprint model of the user for the user is reconstructed in a model adaptation manner;

otherwise, the voiceprint model of the user for the user is reconstructed in a direct re-training manner.

8. The dynamic security code speech-based identity authentication system having self-learning function according to claim 2, wherein

the voiceprint model reconstruction module estimates an updating frequency for updating the parameterized speech data of the user stored in the time-varying data storage unit, and if the updating frequency is less than a predetermined frequency, the voiceprint model of the user for the user is reconstructed in a model adaptation manner;

otherwise, the voiceprint model of the user for the user is reconstructed in a direct re-training manner.

9. The dynamic security code speech-based identity authentication system having self-learning function according to claim 2, wherein

the parameterization module for speech data performs voice activity detection on the speech data before performing a parameterization processing on the speech data.

10. The dynamic security code speech-based identity authentication system having self-learning function according to claim 2, wherein

the parameterization module for speech data uses Mel Frequency Cepstrum Coefficient (MFCC) as a feature parameter, and only saves a zero-order MFCC feature when performing the parameterization processing on the speech data.

11. The dynamic security code speech-based identity authentication system having self-learning function according to claim 1, wherein

the identity authentication module calculates the comprehensive confidence of the identity according to a following equation:

6. (canceled)

12. A dynamic security code speech-based identity authentication method having self-learning function, comprising following steps:

a request receiving step for receiving an identity authentication request that a requester sends to a server through a client;

a dynamic security code generating step for generating a dynamic security code and sending the dynamic security code to the client; and

an identity authentication step for calculating a comprehensive confidence of an identity of the requester by using an acoustic model of global characters and a voiceprint model of a user based on a security code speech signal sent from the client, wherein the security code speech signal is generated when the requester reads out the dynamic security code;

judging the identity of the requester based on the calculated comprehensive confidence of the identity; and

feeding an identity authentication result back to the client,

wherein when the identity authentication result is that the requester is the user of the server, following steps are further performed;

a time-varying data storing step for storing the security code speech signal as a latest speech data into a time-varying data storage unit in which speech data of each user is stored with time labels;

a time window channel construction step for extracting speech data of the user from the time-varying data storage unit in an order of the time labels, constructing a time window channel including a plurality of sets of speech data, and updating the speech data included in the time window channel using the latest speech data; and

a model reconstruction step for reconstructing the voiceprint model of the user for the user using the plurality of sets of speech data included in the updated time window channel.

13. The dynamic security code speech-based identity authentication method having self-learning function according to claim 12, wherein

further comprising a parameterization step, and in the parameterization step, parameterizing the security code speech signal, i.e., speech data, to obtain a latest parameterized speech data;

in the time-varying data storing step, storing the latest parameterized speech data into the time-varying data storage unit in which parameterized speech data of each user is stored with time labels;

in the time window channel construction step, extracting parameterized speech data of the user from the time-varying data storage unit in the order of the time labels, constructing a time window channel including a plurality of sets of parameterized speech data, and updating the parameterized speech data included in the time window channel using the latest parameterized speech data; and

in the model reconstruction step, reconstructing the voiceprint model of the user for the user using the plurality of sets of parameterized speech data included in the updated time window channel.

14. The dynamic security code speech-based identity authentication method having self-learning function according to claim 13, wherein,

further comprising a speech recognition step for recognizing phonemes corresponding to respective frames in the speech data;

in the time-varying data storing step, further storing phonemes corresponding to the latest parameterized speech data and frame intervals corresponding to the phonemes into the time-varying data storage unit; and

in the time window channel construction step, updating the parameterized speech data included in the time window channel based on the phonemes corresponding to the latest parameterized speech data, so that phonemes corresponding to the plurality of sets of parameterized speech data included in the time window channel are evenly distributed.

15. The dynamic security code speech-based identity authentication method having self-learning function according to claim 14, wherein

in the time window channel construction step, comparing the phonemes corresponding to the latest parameterized speech data with phonemes corresponding to an earliest parameterized speech data in the time window channel, and when a matching degree of the phonemes corresponding to the latest parameterized speech data and the phonemes corresponding to the earliest parameterized speech data is greater than or equal to a predetermined matching degree, pushing the latest parameterized speech data into the time window channel, and removing the earliest parameterized speech data from the time window channel.

16. The dynamic security code speech-based identity authentication method having self-learning function according to claim 14, wherein

in the time window channel construction step, tentatively removing a set of parameterized speech data from the time window channel sequentially in the order of the time labels from old to new, and calculating an equilibrium degree of a character-based phoneme distribution based on all of parameterized speech data remained in the time window channel and the latest parameterized speech data, and if the equilibrium degree is greater than or equal to a predetermined threshold of the equilibrium degree, pushing the latest parameterized speech data into the time window channel;

otherwise, restoring the set of parameterized speech data tentatively removed to the time window channel, and then tentatively removing a next set of parameterized speech data from the time window channel, and once again, calculating the equilibrium degree based on all of parameterized speech data remained in the time window channel and the latest parameterized speech data till each set of parameterized speech data included in the time window channel has been tentatively removed or the latest parameterized speech data has been pushed into the time window channel.

18. The dynamic security code speech-based identity authentication method having self-learning function according to claim 12, wherein

in the model reconstruction step, estimating an updating frequency for updating the speech data of the user stored in the time-varying data storage unit, and if the updating frequency is less than a predetermined frequency, reconstructing the voiceprint model of the user for the user in a model adaptation manner;

otherwise, reconstructing the voiceprint model of the user for the user in a direct re-training manner.

19. The dynamic security code speech-based identity authentication method having self-learning function according to claim 13, wherein,

in the model reconstruction step, estimating an updating frequency for updating the parameterized speech data of the user stored in the time-varying data storage unit, and if the updating frequency is less than a predetermined frequency, reconstructing the voiceprint model of the user for the user in a model adaptation manner;

otherwise, reconstructing the voiceprint model of the user for the user in a direct re-training manner.

20. The dynamic security code speech-based identity authentication method having self-learning function according to claim 13, wherein

in the parameterization step, performing voice activity detection on the speech data before performing a parameterization processing on the speech data.

21. The dynamic security code speech-based identity authentication method having self-learning function according to claim 13, wherein

in the parameterization step, using Mel Frequency Cepstrum Coefficient (MFCC) as a feature parameter, and only saving a zero-order MFCC feature when performing the parameterization processing on the speech data.

22. The dynamic security code speech-based identity authentication method having self-learning function according to claim 12, wherein

in the identity authentication step, calculating the comprehensive confidence of the identity according to a following equation:

17. (canceled)